Supplemental material

The folders contain raw hyperspectral images (reported in the paper) of macroalgae sampled in April and July 2024.<code><strong>cropped_HSI_images</strong></code><br>Contains cropped hyperspectral (HSI) raw images prepared for preprocessing. These images are sourced for pixel-wise analysis and further processing steps.<br><code><strong>Optical Signatures</strong></code><br>Houses spectral signatures and labels essential for model training and evaluation. It includes:<code><strong>Train</strong></code><b> Folder</b>: Contains text files of spectral signatures and labels for each preprocessed training sample. These files serve as input to the 1D Convolutional Neural Network (CNN) model, as detailed in the accompanying research paper.<code><strong>Test</strong></code><b> Folder</b>: Holds similar data for the test samples, formatted for model validation and performance evaluation.<br><code><strong>Calibration</strong></code><br>Contains text files of calibration signatures. These are applied in preprocessing for pixel-wise calibration, facilitating the conversion of intensity values to standardized global reflectance.<br><code><strong>Label_data</strong></code><br>Contains bounding box annotations for each sample, crucial for accurate preprocessing and label assignment.<br>The files are named HSI_(30), etc. and correspond to the folder files code names<br><code><strong>Meta_data_sheetmacroalgae_HSI.xlsx</strong></code><br>An Excel file listing species details, name codes, and sample types, aiding in orientation and reference for each dataset file.<br><code><strong>Reviewed_code_manus_hsi</strong></code><br>The file includes a jupyter notebook file with the code for preprocessing the raw images. Analyses of the spectral bands and training a 1D CNN model with that data as well as performing inference on the model. Please see Section 2.3 in the paper for the detailed process.The analysis can be used on your images, but to be sure the code works you are encouraged to use the data presented in the folders here. You can train the model directly without preprocessing first, by using the data in the "Optical signatures"